Zifeng Cheng, Zhonghui Wang, Yuchen Fu, Zhiwei Jiang, Yafeng Yin, Cong Wang, Qing Gu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Yuchen Fu, Zifeng Cheng, Zhiwei Jiang, Zhonghui Wang, Yafeng Yin, Zhengliang Li, Qing Gu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Wireless visual sensor networks (WVSNs) play a prominent role in a wide range of applications such as security, military, and environmental monitoring. As all the visual streaming data from each sensor are required to be delivered to the base station for further processing, this also differs from the data aggregation approaches often used in traditional wireless sensor networks. Therefore, the deployment of WVSNs must not only guarantee the 3D indoor area coverage and connectivity between visual sensors and the base station, but also ensure that the traffic loads are well balanced among the network in order to prolong the lifetime of the entire network. We note that the tree network topology, which is widely used in WSNs, may yield sub-optimal results, since it eliminates the opportunities that some sensors in the network may connect to multiple neighbors closer to the base station. We propose to optimize the traffic load balance by working on the full network topology, transforming it into a generalized max-flow problem and proposing an efficient solution. With this as a building block, we develop a greedy heuristic and an enhanced DFS algorithm to address traffic-aware WVSN deployment problem. Simulations demonstrate that our solutions can significantly reduce the number of sensors while achieving excellent load balance to cover the 3D indoor monitoring space.

Wireless Video Sensor Networks (WVSNs) have been widely proposed for monitoring spatial environments. Dissimilar from traditional Wireless Sensor Networks (WSNs) where nodes such as thermal and light sensors are often considered in 2D and with omni-directional sensing ranges, the sensing range of a video sensor in WVSNs is deemed as directional and usually cognized as a rectangular pyramid shape in 3D environments. Moreover, within indoor spaces, obstacles such as ceiling lights and furniture can easily block the line-of-sight of a video sensor. These new challenges render the traditional deployment solutions for WSNs or 2D modeling environments as impractical to solve the WVSN deployment problem for 3D indoor monitoring. In this paper, we strive to tackle the WVSN deployment problem for 3D indoor monitoring with the consideration of ensuring coverage, connectivity and obstacle awareness. We first model the general problem in a continuous 3D space to minimize the total number of required video sensors. We then convert it into a discrete version problem by incorporating 3D grids, which can achieve arbitrary approximation precision by adjusting the grid granularity. We develop a series of mechanisms to handle the obstacles in the 3D environment and propose an efficient greedy heuristic algorithm yielding high quality solutions. Based on this, we also propose an enhanced Depth First Search (DFS) algorithm that can not only further improve the solution quality, but also return optimal solutions if given enough time. Our extensive simulations demonstrate the superiority of both our greedy heuristic and enhanced DFS solutions.